The argument about AI mastering is usually conducted by two groups talking past each other. One side says a machine cannot replace taste. The other says most masters are a loudness adjustment and a tilt EQ with a mystique attached. Both are partly right, and neither is much help when you have a mix open and a deadline.
A more useful question: what is the master actually for? Answer that, and the choice makes itself.
What mastering is doing
Strip the folklore and mastering does three jobs.
The first is translation. A mix that sounds right in your room has to sound right in a car, on a phone speaker, on cheap earbuds and on a club system. That mostly means tonal balance and controlling the low end.
The second is level. The track needs to sit at a competitive loudness without the distortion, pumping or flattened transients that chasing loudness usually brings.
The third is context. On an album, tracks need to feel like they belong to each other, with sensible gaps and consistent level and tone across the record.
The first two are measurable problems with reference points. The third is a judgement call about a body of work.
What Audial's mastering does
Audial's mastering is one function call. You send a mix, and you get a master back. Optionally you send a reference track as well, and the master is matched toward that reference's loudness and tone.
That reference option is the part worth understanding, because it is where the useful control lives. Without a reference, you are asking for a sensible, competitive master. With one, you are saying "put this in the same territory as that", and "that" can be a commercial release in your genre, another track from the same EP, or a master of your own that you already know translates.
The related analysis function also reports loudness alongside BPM, key, energy and danceability, which makes it easy to check where a mix sits before and after, or to compare a batch of your own tracks to each other.
When AI mastering is enough
Demos, references and client sends. You need the track loud enough to be judged fairly. Nothing about this benefits from a human.
Playlist, library and content work. When a track is one of fifty and nobody will ever listen on a system better than a laptop, consistency beats craft.
Rough mixes during writing. A quick master tells you whether a mix is translating while you can still fix it in the mix, which is where it should be fixed. Mastering a rough mix is diagnostic, not a commitment.
Live recordings, podcasts, sets and streams. The job is level and clarity, delivered quickly, over and over.
Batch work. A hundred tracks that need to land in the same loudness range is exactly the shape of problem an API solves. A human doing that work would be reading a meter.
Genres where the aesthetic is well established. If your reference is a track that already exists and is widely imitated, reference matching gets you close.
When it is not
A record you are releasing to be listened to. Sequencing, the tone relationships between tracks, where the record peaks: that is editorial work about a collection of songs, and an automated master of each individual track will not produce it.
A mix with real problems. If the low end is fighting, a vocal is buried, or there is phase trouble in the drums, automated mastering will make a louder version of those problems. Nothing at the master stage fixes a mix. A good engineer will tell you to go back and fix it, which is itself worth the fee.
Anything where the character is the point. Deliberately lo-fi, deliberately dynamic, deliberately dark records are hard to automate because the processing a system might apply to normalise a track is exactly the thing you were doing on purpose. Automated systems move toward conventional targets, and if your track's identity is a departure from convention, you may hear it being pulled back.
Delivery requirements with teeth. Vinyl cutting, film and broadcast specs, stem delivery, multiple format variants. These come with rules and a person who is accountable for following them.
Stem mastering. When the master needs to change the balance between elements rather than treat the mix as a whole, you need someone working on stems.
When you want an opinion. Part of what you buy from an engineer is a second set of experienced ears telling you what is wrong with your record. That is not a processing service, and no automated system offers it.
A workflow that uses both
The sensible position is not to pick one. It is to use the fast one constantly and the expensive one once.
During writing, master your roughs. Every time you bounce, run it. Listen on your worst speakers. You are checking translation, not polish, and you still have access to the mix.
audial master rough_mix_v4.wav --reference reference_track.wav
Pick your references early and deliberately. Choose two or three commercial tracks that sound like where you want to end up, and use the same ones throughout. This does more for consistency than any processing decision, and it also gives you something concrete to hand an engineer later.
Use automated masters to compare your own mix versions. Mastering flattens the loudness difference between two mixes, which is the single biggest reason A/B comparisons mislead people. Two masters of two mix versions is an honest comparison.
Use it for everything that is not the final release. Demos, sends, promos, socials, library cuts. Charge nobody and wait for nothing.
Send the final to a human if the record warrants it. Give them the mix, the references you used, and the automated master you have been living with. That last item is genuinely useful information: it tells them what you have been hearing and what you expect.
When an engineer is not in the budget, be deliberate instead. Fix the mix until a quick master does not need to work hard. If your automated master sounds like it is straining, the mix is too loud or too unbalanced going in. Back off the mix bus limiting and try again.
The honest summary
Automated mastering made a whole category of mastering work free and instant. That category is real and it is large: everything where the job is translation and level and nobody is making an editorial decision about a body of work.
It did not replace the other category, where you are paying for taste, accountability, and someone willing to tell you your mix is not finished.
Use the first one ten times a week. Use the second one when the record deserves it.
Where to go next
Try it on a mix you already know well, with a reference you already trust, and listen for what it does to the low end and the top.
AI mastering runs in the browser, the mastering function with its optional reference is on the API reference, and plans are on pricing.